Hybrid Optimization of Wheat Yield Using Genetic Algorithms and Machine Learning With Phenotypic and Climatic Features

Abstract

Accurate wheat yield prediction is critical for global food security, yet existing forecasting models often struggle to balance high- dimensional genomic data with dynamic environmental variables. This study developed an automated framework based on genetic algorithms (GAs) to simultaneously optimize phenotypic selection, climatic feature engineering, and machine learning hyper-parameters. The framework was evaluated across two contrasting cultivation environments: irrigated (Mexico) and nonirrigated (Middle East). For the irrigated dataset, the model achieved a peak performance of coefcient of determination (R2) = 0.8363 and root mean squared error (RMSE) = 38.59, demonstrating that the proposed methodology is capable of predicting wheat yield with a R2 exceeding 0.80 under irrigated conditions. Meanwhile, in the nonirrigated environment, the system maintained robust predictive power with R2 = 0.6199 and RMSE = 721.67. To ensure the statistical reliability and reproducibility of these fndings, a bootstrapping validation (1000 iterations) was performed on the top-performing individuals. This process yielded narrow 95% confdence intervals, confrming that while the GA-optimized features provide higher stability in irrigated systems, the framework efectively captures genotype–environment interactions even under water-limited conditions. This dual-environment validation, underpinned by robust resampling techniques, demonstrates the scalability of the proposed soft computing approach for precision breeding across diverse agroclimatic zones.